Again, the releases don't have the libraries.

[0wwafa]

./build/bin/llama-quantize: error while loading shared libraries: libllama.so: cannot open shared object file: No such file or directory

it already happened in the past...

[0wwafa]

Please test the binaries on colab :D if it works there it works everywhere.

[0wwafa]

Version 40xx work..
Version 42xx don't

I tried a few. don't know ehen it "EFFED UP" :D

[dagbs]

Version b4406 built successfully with no libllama.so error.

I setup a script to automatically go through and test until I found one that works. Although, yes... this is currently an issue with latest versions.

To be used with: https://gist.github.com/dagbs/1258a52addeb61cbf26e24f75b2ad7d9

#!/bin/bash

# Starting version and minimum version
start_version=4416
min_version=4000

current_version=$start_version

while [ $current_version -ge $min_version ]; do
    echo "Attempting to build and test version b$current_version"
    
    # Download the specified version
    ./download_latest.sh "b$current_version"
    
    # Build and check if llama-cli works without errors
    if ./build/bin/llama-cli --help 2>&1 | grep -q 'libllama.so'; then
        echo "Error: libllama.so not found in version b$current_version."
    else
        echo "Version b$current_version built successfully with no libllama.so error."
        break # Exit the loop if a successful build is found
    fi
    
    # Decrease the version number and wait 5 seconds before the next attempt
    current_version=$((current_version - 1))
    sleep 5
done

if [ $current_version -lt $min_version ]; then
    echo "No suitable version found. All versions from b$start_version to b$min_version had issues."
fi

[bandoti]

@0wwafa Do you mean to say that when downloading llama.cpp it is not including the shared libraries? If so where are you downloading from? Or are you building, installing, and then running the application and those shared libraries are missing?

If this happened before it is helpful to include a link to a previous issue or PR that resolved the problem you are seeing.

Folks use llama.cpp in various ways so it is necessary to have the extra context so we can help resolve the issue accordingly!

[pjh64]

I ran into the same issue and I remember it to also happen on other platforms such as x86_64.
There seem to be two inconveniences for me. The first is, that cmake does seem to be necessary now. I really appreciated the ease of just running make on even the most basic systems to compile llama.cpp. But that aside, this change will certainly have its reasons. The library issue however seems to be much more of a problem.
It works however when calling the binaries directly from their build dir.

Here is the entire build log from my last compiling attempt on the arm64 platform (but the same happens on x86_64):
:
https://pastebin.com/EK6vNMn0